Generating explanations of machine learning predictions in matching problems

Abstract

Methods, systems, and computer-readable storage media for receiving a set of documents matched by a ML model, each document in the set of documents including a computer-readable electronic document, processing a set of pairwise features, the ML model, and the set of documents by an explanation framework to generate a set of raw explanations, the set of raw explanations including one or more raw explanations, each raw explanation including a pairwise feature and an importance score, for each raw explanation, identifying a natural language template based on the pairwise feature and the importance score, and populating the natural language template with one or more parameters provided from the set of documents to provide a natural language explanation for matching of the documents in the set of documents by the ML model.

What is claimed is: 
     
       1. A computer-implemented method for providing natural language explanations for document matches of machine learning (ML) models, the method being executed by one or more processors and comprising:
 receiving a set of documents matched by a ML model, each document in the set of documents comprising a computer-readable electronic document; 
 processing a set of pairwise features, the ML model, and the set of documents by an explanation framework to generate a set of raw explanations, the set of raw explanations comprising one or more raw explanations, each raw explanation comprising a pairwise feature and an importance score; 
 for each raw explanation, identifying a natural language template based on the pairwise feature and the importance score; and 
 populating the natural language template with one or more parameters provided from the set of documents to provide a natural language explanation for matching of the documents in the set of documents by the ML model. 
 
     
     
       2. The method of  claim 1 , wherein identifying a natural language template based on the pairwise feature and the importance score comprises:
 determining a set of natural language templates based on the pairwise feature; and 
 selecting the natural language template from the set of natural language templates based on the importance score. 
 
     
     
       3. The method of  claim 2 , wherein determining a set of natural language templates based on the pairwise feature comprises identifying a feature code for the pairwise feature, and identifying the set of natural language templates based on the feature code. 
     
     
       4. The method of  claim 1 , further comprising determining a feature descriptor for the set of documents, the feature descriptor comprising a set of pairwise features provided by processing features based on binary operators. 
     
     
       5. The method of  claim 1 , wherein each parameter comprises a value determined from a document in the set of documents. 
     
     
       6. The method of  claim 1 , wherein the explanation framework randomly perturbates input to the ML model to affect predictions of the ML model and generate an importance score for each pairwise feature. 
     
     
       7. The method of  claim 1 , wherein the set of documents comprise a bank statement and an invoice. 
     
     
       8. A non-transitory computer-readable storage medium coupled to one or more processors and having instructions stored thereon which, when executed by the one or more processors, cause the one or more processors to perform operations for natural language explanations for document matches of machine learning (ML) models, the operations comprising:
 receiving a set of documents matched by a ML model, each document in the set of documents comprising a computer-readable electronic document; 
 processing a set of pairwise features, the ML model, and the set of documents by an explanation framework to generate a set of raw explanations, the set of raw explanations comprising one or more raw explanations, each raw explanation comprising a pairwise feature and an importance score; 
 for each raw explanation, identifying a natural language template based on the pairwise feature and the importance score; and 
 populating the natural language template with one or more parameters provided from the set of documents to provide a natural language explanation for matching of the documents in the set of documents by the ML model. 
 
     
     
       9. The computer-readable storage medium of  claim 8 , wherein identifying a natural language template based on the pairwise feature and the importance score comprises:
 determining a set of natural language templates based on the pairwise feature; and 
 selecting the natural language template from the set of natural language templates based on the importance score. 
 
     
     
       10. The computer-readable storage medium of  claim 9 , wherein determining a set of natural language templates based on the pairwise feature comprises identifying a feature code for the pairwise feature, and identifying the set of natural language templates based on the feature code. 
     
     
       11. The computer-readable storage medium of  claim 8 , wherein operations further comprise determining a feature descriptor for the set of documents, the feature descriptor comprising a set of pairwise features provided by processing features based on binary operators. 
     
     
       12. The computer-readable storage medium of  claim 8 , wherein each parameter comprises a value determined from a document in the set of documents. 
     
     
       13. The computer-readable storage medium of  claim 8 , wherein the explanation framework randomly perturbates input to the ML model to affect predictions of the ML model and generate an importance score for each pairwise feature. 
     
     
       14. The computer-readable storage medium of  claim 8 , wherein the set of documents comprise a bank statement and an invoice. 
     
     
       15. A system, comprising:
 a computing device; and 
 a computer-readable storage device coupled to the computing device and having instructions stored thereon which, when executed by the computing device, cause the computing device to perform operations for natural language explanations for document matches of machine learning (ML) models, the operations comprising:
 receiving a set of documents matched by a ML model, each document in the set of documents comprising a computer-readable electronic document; 
 processing a set of pairwise features, the ML model, and the set of documents by an explanation framework to generate a set of raw explanations, the set of raw explanations comprising one or more raw explanations, each raw explanation comprising a pairwise feature and an importance score; 
 for each raw explanation, identifying a natural language template based on the pairwise feature and the importance score; and 
 populating the natural language template with one or more parameters provided from the set of documents to provide a natural language explanation for matching of the documents in the set of documents by the ML model. 
 
 
     
     
       16. The system of  claim 15 , wherein identifying a natural language template based on the pairwise feature and the importance score comprises:
 determining a set of natural language templates based on the pairwise feature; and 
 selecting the natural language template from the set of natural language templates based on the importance score. 
 
     
     
       17. The system of  claim 16 , wherein determining a set of natural language templates based on the pairwise feature comprises identifying a feature code for the pairwise feature, and identifying the set of natural language templates based on the feature code. 
     
     
       18. The system of  claim 15 , wherein operations further comprise determining a feature descriptor for the set of documents, the feature descriptor comprising a set of pairwise features provided by processing features based on binary operators. 
     
     
       19. The system of  claim 15 , wherein each parameter comprises a value determined from a document in the set of documents. 
     
     
       20. The system of  claim 15 , wherein the explanation framework randomly perturbates input to the ML model to affect predictions of the ML model and generate an importance score for each pairwise feature.